OpenAI has launched AI Futures, the blog of its new Strategic Futures team. The team’s stated mission is to answer a single broad question: how should free society be restructured to preserve individual rights and agency while accommodating the emergence of transformative AI? The post frames this as a concentration-of-power risk, arguing that this category is the largest, most serious, and most conceptually challenging issue in AI safety and policy.

The argument begins with a historical observation: political and military power has always depended on the large-scale cooperation and consent of human beings. States project force through soldiers and police, fund themselves through taxes on human labor, and run bureaucracies staffed by people. The post cites David Hume’s remark that political leaders have nothing to support them but popular opinion. Advances in autonomous systems and machine intelligence could break that dependency. States may soon project force without cooperative soldiers or police, collect revenue from data-center outputs rather than human labor, and automate much of the bureaucracy. If so, power would no longer require a society-wide bargain, and ordinary people could lose their seat at the negotiating table.

The post invokes James Madison and the Federalist Papers to argue that parchment barriers alone cannot prevent tyranny. The Founders, it says, designed checks and balances using Newtonian metaphors of orbits and attraction, aiming not for radical decentralization but for a balance of power that checks ambition against ambition. The Strategic Futures team applies this logic to AI: the goal is not maximum decentralization but the right balance, such that no single actor or small set of actors can dominate the rest. That balance must avoid both a world where any malicious individual can cause mass harm and a world where individuals lack broad control over the tools they use to express freedom. The post also notes risks from non-malicious actors, citing a Hugging Face incident where AI agents acted beyond their assigned tasks, and raises concerns about superintelligent AI untethered from human control.

The team outlines a non-exhaustive set of guiding principles: humans should maintain individual autonomy and opportunity in AI use even at the expense of security and economic growth; individual autonomy depends on individual responsibility for misuse; a small category of serious risks requires collective action, but that action should be as narrow and modest as possible; law should empower individuals and small organizations rather than centralize power; human institutions should retain primacy in world affairs; and long-run governance requires new institutional mechanisms for bounded legibility—tying high-stakes AI actions back to a responsible human or organization—while keeping privacy and anonymity at the core.

The post acknowledges that these principles exist in tension and that the work will require research across public-policy design, economics, law, history, and machine learning, as well as forecasting how AI might transform firms and government agencies, much as railroads gave rise to the modern managerial corporation. The Strategic Futures team plans to publish papers, videos, and podcasts, and will emphasize iteration, debate, and receptivity to feedback. It concludes that no single company or society can answer these questions unilaterally; the task is collective, and the work has just begun.